MiniMax H3 · Delivery Checkpoint → Editor Chunk

The IAMCCS shotboard workflow doesn't stop at rendering - a rendered MiniMax H3 shot is written to disk as an MP4, and then the in-graph Shotboard Video Editor wants to show it on a timeline lane. Between the two sits this node. It takes the path of the rendered delivery file, decodes it back into IMAGE frames plus AUDIO, and hands them to the editor along with a readiness flag. It's the "delivery checkpoint → editor chunk" bridge, which is basically the subtitle of the node.

The interesting behavior is how it treats long-form vs. per-shot rendering:

• LongVid mode: every shot renders into one film. Early segments return the native frames with master_ready = false and a "waiting for LongVid master" report; only when the final segment completes does it decode the finished final_film.mp4 and report master_ready = true. The editor gets one coherent asset, not a pile of partial chunks.
• Per-shot modes (each chunk is its own roll - e.g. LTX 2.3 per-chunk editorial work): it resolves the deterministic segment_XXXX.mp4 for the current shot so the editor always sees the right roll on the right lane, even on the final pass when the delivery node also wrote a combined film.

How it works

• It checks the path exists (it decodes from disk, so the checkpoint has to be real - hence it forces a rerun via IS_CHANGED returning NaN rather than trusting cached readiness).
• Decodes through the editor's video input implementation, grabbing frames and audio.
• Falls back to the passed-in native_audio if the file's audio track isn't usable.
• If the resolved file is missing, it raises FileNotFoundError naming the exact path it looked for - usually the fastest way to discover the delivery node never actually wrote where you think.

Inputs

Required: cine_linx (shot plan), delivery_video_path (forceInput, the path from the delivery/checkpoint node), current_segment, total_segments, native_frames, native_audio. The native frames/audio act as the standing material while the real delivery is still writing.

Outputs

• delivery_frames, delivery_audio - the decoded clip for the editor.
• master_ready - boolean the editor watches.
• resolved_path - the exact file that was decoded.
• report - delivery kind (LongVid master vs. editorial roll vs. slot), frame count, fps, and path.

Install

Pack install (git clone https://github.com/IAMCCS/IAMCCS-nodes.git), MiniMax H3 support, plus whatever the editor delivery path uses to read video files in your ComfyUI build (the decoder is ComfyUI's own).

Gotchas

• If the editor shows native frames forever, check master_ready: in LongVid mode this node deliberately holds back the real master until the last segment. It's not broken; the film isn't done.
• Missing checkpoint errors almost always mean the render wrote somewhere else or the run was interrupted - read the resolved_path in the error and compare with what's on disk.
• This node reads MP4s but doesn't save them. The saving is done upstream by the delivery/checkpoint output stages; this one is the reader side.
• It's deep editor plumbing: without the IAMCCS Shotboard Editor lanes connected, it's a decoder with extra steps.